datapath/datapath.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008 The Board of Trustees of The Leland
   4  * Stanford Junior University
   5  */
   6
   7 /* Functions for managing the dp interface/device. */
   8
   9 #include <linux/init.h>
  10 #include <linux/module.h>
  11 #include <linux/if_arp.h>
  12 #include <linux/if_bridge.h>
  13 #include <linux/if_vlan.h>
  14 #include <linux/in.h>
  15 #include <net/genetlink.h>
  16 #include <linux/ip.h>
  17 #include <linux/delay.h>
  18 #include <linux/etherdevice.h>
  19 #include <linux/kernel.h>
  20 #include <linux/kthread.h>
  21 #include <linux/mutex.h>
  22 #include <linux/rtnetlink.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/version.h>
  25 #include <linux/ethtool.h>
  26 #include <linux/random.h>
  27 #include <asm/system.h>
  28 #include <linux/netfilter_bridge.h>
  29 #include <linux/netfilter_ipv4.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/list.h>
  32 #include <linux/rculist.h>
  33 #include <linux/workqueue.h>
  34
  35 #include "openflow/openflow-netlink.h"
  36 #include "datapath.h"
  37 #include "nx_act_snat.h"
  38 #include "table.h"
  39 #include "chain.h"
  40 #include "dp_dev.h"
  41 #include "forward.h"
  42 #include "flow.h"
  43
  44 #include "compat.h"
  45
  46
  47 /* Strings to describe the manufacturer, hardware, and software.  This data
  48  * is queriable through the switch description stats message. */
  49 static char mfr_desc[DESC_STR_LEN] = "Nicira Networks";
  50 static char hw_desc[DESC_STR_LEN] = "Reference Linux Kernel Module";
  51 static char sw_desc[DESC_STR_LEN] = VERSION;
  52 static char serial_num[SERIAL_NUM_LEN] = "None";
  53
  54 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  55 module_param_string(mfr_desc, mfr_desc, sizeof mfr_desc, 0444);
  56 module_param_string(hw_desc, hw_desc, sizeof hw_desc, 0444);
  57 module_param_string(sw_desc, sw_desc, sizeof sw_desc, 0444);
  58 module_param_string(serial_num, serial_num, sizeof serial_num, 0444);
  59 #else
  60 MODULE_PARM(mfr_desc, "s");
  61 MODULE_PARM(hw_desc, "s");
  62 MODULE_PARM(sw_desc, "s");
  63 MODULE_PARM(serial_num, "s");
  64 #endif
  65
  66
  67 /* Number of milliseconds between runs of the maintenance thread. */
  68 #define MAINT_SLEEP_MSECS 1000
  69
  70 #define UINT32_MAX                        4294967295U
  71 #define UINT16_MAX                        65535
  72 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
  73
  74 static struct genl_family dp_genl_family;
  75 static struct genl_multicast_group mc_group;
  76
  77 /* It's hard to imagine wanting more than one datapath, but... */
  78 #define DP_MAX 32
  79
  80 /* Datapaths.  Protected on the read side by rcu_read_lock, on the write side
  81  * by dp_mutex.  dp_mutex is almost completely redundant with genl_mutex
  82  * maintained by the Generic Netlink code, but the timeout path needs mutual
  83  * exclusion too.
  84  *
  85  * It is safe to access the datapath and net_bridge_port structures with just
  86  * dp_mutex.
  87  */
  88 static struct datapath *dps[DP_MAX];
  89 DEFINE_MUTEX(dp_mutex);
  90 EXPORT_SYMBOL(dp_mutex);
  91
  92 static int dp_maint_func(void *data);
  93 static void init_port_status(struct net_bridge_port *p);
  94 static int dp_genl_openflow_done(struct netlink_callback *);
  95 static struct net_bridge_port *new_nbp(struct datapath *,
  96                                        struct net_device *, int port_no);
  97
  98 /* nla_shrink - reduce amount of space reserved by nla_reserve
  99  * @skb: socket buffer from which to recover room
 100  * @nla: netlink attribute to adjust
 101  * @len: new length of attribute payload
 102  *
 103  * Reduces amount of space reserved by a call to nla_reserve.
 104  *
 105  * No other attributes may be added between calling nla_reserve and this
 106  * function, since it will create a hole in the message.
 107  */
 108 void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
 109 {
 110         int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
 111         BUG_ON(delta > 0);
 112         skb->tail += delta;
 113         skb->len  += delta;
 114         nla->nla_len = nla_attr_size(len);
 115 }
 116
 117 /* Puts a set of openflow headers for a message of the given 'type' into 'skb'.
 118  * If 'sender' is nonnull, then it is used as the message's destination.  'dp'
 119  * must specify the datapath to use.
 120  *
 121  * '*max_openflow_len' receives the maximum number of bytes that are available
 122  * for the embedded OpenFlow message.  The caller must call
 123  * resize_openflow_skb() to set the actual size of the message to this number
 124  * of bytes or less.
 125  *
 126  * Returns the openflow header if successful, otherwise (if 'skb' is too small)
 127  * an error code. */
 128 static void *
 129 put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
 130                      const struct sender *sender, int *max_openflow_len)
 131 {
 132         struct ofp_header *oh;
 133         struct nlattr *attr;
 134         int openflow_len;
 135
 136         /* Assemble the Generic Netlink wrapper. */
 137         if (!genlmsg_put(skb,
 138                          sender ? sender->pid : 0,
 139                          sender ? sender->seq : 0,
 140                          &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
 141                 return ERR_PTR(-ENOBUFS);
 142         if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
 143                 return ERR_PTR(-ENOBUFS);
 144         openflow_len = (skb_tailroom(skb) - NLA_HDRLEN) & ~(NLA_ALIGNTO - 1);
 145         if (openflow_len < sizeof *oh)
 146                 return ERR_PTR(-ENOBUFS);
 147         *max_openflow_len = openflow_len;
 148         attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
 149         BUG_ON(!attr);
 150
 151         /* Fill in the header.  The caller is responsible for the length. */
 152         oh = nla_data(attr);
 153         oh->version = OFP_VERSION;
 154         oh->type = type;
 155         oh->xid = sender ? sender->xid : 0;
 156
 157         return oh;
 158 }
 159
 160 /* Resizes OpenFlow header 'oh', which must be at the tail end of 'skb', to new
 161  * length 'new_length' (in bytes), adjusting pointers and size values as
 162  * necessary. */
 163 static void
 164 resize_openflow_skb(struct sk_buff *skb,
 165                     struct ofp_header *oh, size_t new_length)
 166 {
 167         struct nlattr *attr = ((void *) oh) - NLA_HDRLEN;
 168         nla_shrink(skb, attr, new_length);
 169         oh->length = htons(new_length);
 170         nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 171 }
 172
 173 /* Allocates a new skb to contain an OpenFlow message 'openflow_len' bytes in
 174  * length.  Returns a null pointer if memory is unavailable, otherwise returns
 175  * the OpenFlow header and stores a pointer to the skb in '*pskb'.
 176  *
 177  * 'type' is the OpenFlow message type.  If 'sender' is nonnull, then it is
 178  * used as the message's destination.  'dp' must specify the datapath to
 179  * use.  */
 180 static void *
 181 alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
 182                    const struct sender *sender, struct sk_buff **pskb)
 183 {
 184         struct ofp_header *oh;
 185         size_t genl_len;
 186         struct sk_buff *skb;
 187         int max_openflow_len;
 188
 189         if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
 190                 if (net_ratelimit())
 191                         printk("alloc_openflow_skb: openflow message too large: %zu\n",
 192                                         openflow_len);
 193                 return NULL;
 194         }
 195
 196         genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
 197         genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
 198         genl_len += nla_total_size(openflow_len);    /* DP_GENL_A_OPENFLOW */
 199         skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
 200         if (!skb) {
 201                 if (net_ratelimit())
 202                         printk("alloc_openflow_skb: genlmsg_new failed\n");
 203                 return NULL;
 204         }
 205
 206         oh = put_openflow_headers(dp, skb, type, sender, &max_openflow_len);
 207         BUG_ON(!oh || IS_ERR(oh));
 208         resize_openflow_skb(skb, oh, openflow_len);
 209
 210         return oh;
 211 }
 212
 213 /* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
 214  * listeners. */
 215 static int
 216 send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
 217 {
 218         return (sender
 219                 ? genlmsg_unicast(skb, sender->pid)
 220                 : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
 221 }
 222
 223 /* Generates a unique datapath id.  It incorporates the datapath index
 224  * and a hardware address, if available.  If not, it generates a random
 225  * one.
 226  */
 227 static
 228 uint64_t gen_datapath_id(uint16_t dp_idx)
 229 {
 230         uint64_t id;
 231         int i;
 232         struct net_device *dev;
 233
 234         /* The top 16 bits are used to identify the datapath.  The lower 48 bits
 235          * use an interface address.  */
 236         id = (uint64_t)dp_idx << 48;
 237         if ((dev = dev_get_by_name(&init_net, "ctl0"))
 238                         || (dev = dev_get_by_name(&init_net, "eth0"))) {
 239                 for (i=0; i<ETH_ALEN; i++) {
 240                         id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
 241                 }
 242                 dev_put(dev);
 243         } else {
 244                 /* Randomly choose the lower 48 bits if we cannot find an
 245                  * address and mark the most significant bit to indicate that
 246                  * this was randomly generated. */
 247                 uint8_t rand[ETH_ALEN];
 248                 get_random_bytes(rand, ETH_ALEN);
 249                 id |= (uint64_t)1 << 63;
 250                 for (i=0; i<ETH_ALEN; i++) {
 251                         id |= (uint64_t)rand[i] << (8*(ETH_ALEN-1 - i));
 252                 }
 253         }
 254
 255         return id;
 256 }
 257
 258 /* Creates a new datapath numbered 'dp_idx'.  Returns 0 for success or a
 259  * negative error code. */
 260 static int new_dp(int dp_idx)
 261 {
 262         struct datapath *dp;
 263         int err;
 264
 265         if (dp_idx < 0 || dp_idx >= DP_MAX)
 266                 return -EINVAL;
 267
 268         if (!try_module_get(THIS_MODULE))
 269                 return -ENODEV;
 270
 271         /* Exit early if a datapath with that number already exists. */
 272         if (dps[dp_idx]) {
 273                 err = -EEXIST;
 274                 goto err_unlock;
 275         }
 276
 277         err = -ENOMEM;
 278         dp = kzalloc(sizeof *dp, GFP_KERNEL);
 279         if (dp == NULL)
 280                 goto err_unlock;
 281
 282         /* Setup our "of" device */
 283         err = dp_dev_setup(dp);
 284         if (err)
 285                 goto err_free_dp;
 286
 287         dp->dp_idx = dp_idx;
 288         dp->id = gen_datapath_id(dp_idx);
 289         dp->chain = chain_create(dp);
 290         if (dp->chain == NULL)
 291                 goto err_destroy_dp_dev;
 292         INIT_LIST_HEAD(&dp->port_list);
 293
 294         dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
 295         if (IS_ERR(dp->local_port)) {
 296                 err = PTR_ERR(dp->local_port);
 297                 goto err_destroy_local_port;
 298         }
 299
 300         dp->flags = 0;
 301         dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
 302
 303         dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
 304         if (IS_ERR(dp->dp_task))
 305                 goto err_destroy_chain;
 306
 307         dps[dp_idx] = dp;
 308
 309         return 0;
 310
 311 err_destroy_local_port:
 312         dp_del_switch_port(dp->local_port);
 313 err_destroy_chain:
 314         chain_destroy(dp->chain);
 315 err_destroy_dp_dev:
 316         dp_dev_destroy(dp);
 317 err_free_dp:
 318         kfree(dp);
 319 err_unlock:
 320         module_put(THIS_MODULE);
 321                 return err;
 322 }
 323
 324 /* Find and return a free port number under 'dp'. */
 325 static int find_portno(struct datapath *dp)
 326 {
 327         int i;
 328         for (i = 0; i < DP_MAX_PORTS; i++)
 329                 if (dp->ports[i] == NULL)
 330                         return i;
 331         return -EXFULL;
 332 }
 333
 334 static struct net_bridge_port *new_nbp(struct datapath *dp,
 335                                        struct net_device *dev, int port_no)
 336 {
 337         struct net_bridge_port *p;
 338
 339         if (dev->br_port != NULL)
 340                 return ERR_PTR(-EBUSY);
 341
 342         p = kzalloc(sizeof(*p), GFP_KERNEL);
 343         if (p == NULL)
 344                 return ERR_PTR(-ENOMEM);
 345
 346         rtnl_lock();
 347         dev_set_promiscuity(dev, 1);
 348         rtnl_unlock();
 349         dev_hold(dev);
 350         p->dp = dp;
 351         p->dev = dev;
 352         p->port_no = port_no;
 353         spin_lock_init(&p->lock);
 354         INIT_WORK(&p->port_task, NULL);
 355         if (port_no != OFPP_LOCAL)
 356                 rcu_assign_pointer(dev->br_port, p);
 357         if (port_no < DP_MAX_PORTS)
 358                 rcu_assign_pointer(dp->ports[port_no], p);
 359         list_add_rcu(&p->node, &dp->port_list);
 360
 361         return p;
 362 }
 363
 364 int add_switch_port(struct datapath *dp, struct net_device *dev)
 365 {
 366         struct net_bridge_port *p;
 367         int port_no;
 368
 369         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
 370             || is_dp_dev(dev))
 371                 return -EINVAL;
 372
 373         port_no = find_portno(dp);
 374         if (port_no < 0)
 375                 return port_no;
 376
 377         p = new_nbp(dp, dev, port_no);
 378         if (IS_ERR(p))
 379                 return PTR_ERR(p);
 380
 381         init_port_status(p);
 382
 383         /* Notify the ctlpath that this port has been added */
 384         dp_send_port_status(p, OFPPR_ADD);
 385
 386         return 0;
 387 }
 388
 389 /* Delete 'p' from switch. */
 390 int dp_del_switch_port(struct net_bridge_port *p)
 391 {
 392 #ifdef SUPPORT_SNAT
 393         unsigned long flags;
 394 #endif
 395
 396         /* First drop references to device. */
 397         cancel_work_sync(&p->port_task);
 398         rtnl_lock();
 399         dev_set_promiscuity(p->dev, -1);
 400         rtnl_unlock();
 401         list_del_rcu(&p->node);
 402         if (p->port_no != OFPP_LOCAL)
 403                 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 404         rcu_assign_pointer(p->dev->br_port, NULL);
 405
 406         /* Then wait until no one is still using it, and destroy it. */
 407         synchronize_rcu();
 408
 409 #ifdef SUPPORT_SNAT
 410         /* Free any SNAT configuration on the port. */
 411         spin_lock_irqsave(&p->lock, flags);
 412         snat_free_conf(p);
 413         spin_unlock_irqrestore(&p->lock, flags);
 414 #endif
 415
 416         /* Notify the ctlpath that this port no longer exists */
 417         dp_send_port_status(p, OFPPR_DELETE);
 418
 419         dev_put(p->dev);
 420         kfree(p);
 421
 422         return 0;
 423 }
 424
 425 static void del_dp(struct datapath *dp)
 426 {
 427         struct net_bridge_port *p, *n;
 428
 429         kthread_stop(dp->dp_task);
 430
 431         /* Drop references to DP. */
 432         list_for_each_entry_safe (p, n, &dp->port_list, node)
 433                 dp_del_switch_port(p);
 434         rcu_assign_pointer(dps[dp->dp_idx], NULL);
 435
 436         /* Kill off local_port dev references from buffered packets that have
 437          * associated dst entries. */
 438         synchronize_rcu();
 439         fwd_discard_all();
 440
 441         /* Destroy dp->netdev.  (Must follow deleting switch ports since
 442          * dp->local_port has a reference to it.) */
 443         dp_dev_destroy(dp);
 444
 445         /* Wait until no longer in use, then destroy it. */
 446         synchronize_rcu();
 447         chain_destroy(dp->chain);
 448         kfree(dp);
 449         module_put(THIS_MODULE);
 450 }
 451
 452 static int dp_maint_func(void *data)
 453 {
 454         struct datapath *dp = (struct datapath *) data;
 455
 456         while (!kthread_should_stop()) {
 457 #ifdef SUPPORT_SNAT
 458                 struct net_bridge_port *p;
 459
 460                 /* Expire old SNAT entries */
 461                 rcu_read_lock();
 462                 list_for_each_entry_rcu (p, &dp->port_list, node)
 463                         snat_maint(p);
 464                 rcu_read_unlock();
 465 #endif
 466
 467                 /* Timeout old entries */
 468                 chain_timeout(dp->chain);
 469                 msleep_interruptible(MAINT_SLEEP_MSECS);
 470         }
 471
 472         return 0;
 473 }
 474
 475 static void
 476 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
 477 {
 478 #ifdef SUPPORT_SNAT
 479         /* Check if this packet needs early SNAT processing. */
 480         if (snat_pre_route(skb)) {
 481                 kfree_skb(skb);
 482                 return;
 483         }
 484 #endif
 485
 486         /* Push the Ethernet header back on. */
 487         skb_push(skb, ETH_HLEN);
 488         fwd_port_input(p->dp->chain, skb, p);
 489 }
 490
 491 /*
 492  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
 493  * different set of devices!)
 494  */
 495 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
 496 /* Called with rcu_read_lock. */
 497 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
 498                                          struct sk_buff *skb)
 499 {
 500         do_port_input(p, skb);
 501         return NULL;
 502 }
 503 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
 504 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 505 {
 506         do_port_input(p, *pskb);
 507         return 1;
 508 }
 509 #else
 510 /* NB: This has only been tested on 2.4.35 */
 511 static void dp_frame_hook(struct sk_buff *skb)
 512 {
 513         struct net_bridge_port *p = skb->dev->br_port;
 514         if (p) {
 515                 rcu_read_lock();
 516                 do_port_input(p, skb);
 517                 rcu_read_unlock();
 518         } else
 519                 kfree_skb(skb);
 520 }
 521 #endif
 522
 523 /* Forwarding output path.
 524  * Based on net/bridge/br_forward.c. */
 525
 526 static inline unsigned packet_length(const struct sk_buff *skb)
 527 {
 528         int length = skb->len - ETH_HLEN;
 529         if (skb->protocol == htons(ETH_P_8021Q))
 530                 length -= VLAN_HLEN;
 531         return length;
 532 }
 533
 534 /* Send packets out all the ports except the originating one.  If the
 535  * "flood" argument is set, only send along the minimum spanning tree.
 536  */
 537 static int
 538 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
 539 {
 540         u32 disable = flood ? OFPPC_NO_FLOOD : 0;
 541         struct net_bridge_port *p;
 542         int prev_port = -1;
 543
 544         list_for_each_entry_rcu (p, &dp->port_list, node) {
 545                 if (skb->dev == p->dev || p->config & disable)
 546                         continue;
 547                 if (prev_port != -1) {
 548                         struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
 549                         if (!clone) {
 550                                 kfree_skb(skb);
 551                                 return -ENOMEM;
 552                         }
 553                         dp_output_port(dp, clone, prev_port, 0);
 554                 }
 555                 prev_port = p->port_no;
 556         }
 557         if (prev_port != -1)
 558                 dp_output_port(dp, skb, prev_port, 0);
 559         else
 560                 kfree_skb(skb);
 561
 562         return 0;
 563 }
 564
 565 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
 566    FIXME: how are devices reference counted? */
 567 void dp_set_origin(struct datapath *dp, uint16_t in_port,
 568                            struct sk_buff *skb)
 569 {
 570         struct net_bridge_port *p;
 571         p = (in_port < DP_MAX_PORTS ? dp->ports[in_port]
 572              : in_port == OFPP_LOCAL ? dp->local_port
 573              : NULL);
 574         if (p)
 575                 skb->dev = p->dev;
 576          else
 577                 skb->dev = NULL;
 578 }
 579
 580 int
 581 dp_xmit_skb(struct sk_buff *skb)
 582 {
 583         int len = skb->len;
 584         if (packet_length(skb) > skb->dev->mtu) {
 585                 printk("dropped over-mtu packet: %d > %d\n",
 586                            packet_length(skb), skb->dev->mtu);
 587                 kfree_skb(skb);
 588                 return -E2BIG;
 589         }
 590
 591         dev_queue_xmit(skb);
 592
 593         return len;
 594 }
 595
 596 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
 597  */
 598 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
 599                    int ignore_no_fwd)
 600 {
 601         BUG_ON(!skb);
 602         switch (out_port){
 603         case OFPP_IN_PORT:
 604                 /* Send it out the port it came in on, which is already set in
 605                  * the skb. */
 606                 if (!skb->dev) {
 607                         if (net_ratelimit())
 608                                 printk("skb device not set forwarding to in_port\n");
 609                         kfree_skb(skb);
 610                         return -ESRCH;
 611                 }
 612                 return dp_xmit_skb(skb);
 613
 614         case OFPP_TABLE: {
 615                 int retval = run_flow_through_tables(dp->chain, skb,
 616                                                      skb->dev->br_port);
 617                 if (retval)
 618                         kfree_skb(skb);
 619                 return retval;
 620         }
 621
 622         case OFPP_FLOOD:
 623                 return output_all(dp, skb, 1);
 624
 625         case OFPP_ALL:
 626                 return output_all(dp, skb, 0);
 627
 628         case OFPP_CONTROLLER:
 629                 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
 630                                                   OFPR_ACTION);
 631
 632         case OFPP_LOCAL: {
 633                 struct net_device *dev = dp->netdev;
 634 #ifdef SUPPORT_SNAT
 635                 snat_local_in(skb);
 636 #endif
 637                 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
 638         }
 639
 640         case 0 ... DP_MAX_PORTS - 1: {
 641                 struct net_bridge_port *p = dp->ports[out_port];
 642                 if (p == NULL)
 643                         goto bad_port;
 644                 if (p->dev == skb->dev) {
 645                         /* To send to the input port, must use OFPP_IN_PORT */
 646                         kfree_skb(skb);
 647                         if (net_ratelimit())
 648                                 printk("can't directly forward to input port\n");
 649                         return -EINVAL;
 650                 }
 651                 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
 652                         kfree_skb(skb);
 653                         return 0;
 654                 }
 655                 skb->dev = p->dev;
 656                 return dp_xmit_skb(skb);
 657         }
 658
 659         default:
 660                 goto bad_port;
 661         }
 662
 663 bad_port:
 664         kfree_skb(skb);
 665         if (net_ratelimit())
 666                 printk("can't forward to bad port %d\n", out_port);
 667         return -ENOENT;
 668 }
 669
 670 /* Takes ownership of 'skb' and transmits it to 'dp''s control path.  If
 671  * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
 672  * otherwise, all of 'skb' is sent.  'reason' indicates why 'skb' is being
 673  * sent. 'max_len' sets the maximum number of bytes that the caller
 674  * wants to be sent; a value of 0 indicates the entire packet should be
 675  * sent. */
 676 int
 677 dp_output_control(struct datapath *dp, struct sk_buff *skb,
 678                            uint32_t buffer_id, size_t max_len, int reason)
 679 {
 680         /* FIXME?  Can we avoid creating a new skbuff in the case where we
 681          * forward the whole packet? */
 682         struct sk_buff *f_skb;
 683         struct ofp_packet_in *opi;
 684         struct net_bridge_port *p;
 685         size_t fwd_len, opi_len;
 686         int err;
 687
 688         fwd_len = skb->len;
 689         if ((buffer_id != (uint32_t) -1) && max_len)
 690                 fwd_len = min(fwd_len, max_len);
 691
 692         opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
 693         opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
 694         if (!opi) {
 695                 err = -ENOMEM;
 696                 goto out;
 697         }
 698         opi->buffer_id      = htonl(buffer_id);
 699         opi->total_len      = htons(skb->len);
 700         p = skb->dev->br_port;
 701         opi->in_port        = htons(p ? p->port_no : OFPP_LOCAL);
 702         opi->reason         = reason;
 703         opi->pad            = 0;
 704         memcpy(opi->data, skb_mac_header(skb), fwd_len);
 705         err = send_openflow_skb(f_skb, NULL);
 706
 707 out:
 708         kfree_skb(skb);
 709         return err;
 710 }
 711
 712 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
 713 {
 714         unsigned long flags;
 715         desc->port_no = htons(p->port_no);
 716         strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
 717         desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
 718         memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
 719         desc->curr = 0;
 720         desc->supported = 0;
 721         desc->advertised = 0;
 722         desc->peer = 0;
 723
 724         spin_lock_irqsave(&p->lock, flags);
 725         desc->config = htonl(p->config);
 726         desc->state = htonl(p->state);
 727         spin_unlock_irqrestore(&p->lock, flags);
 728
 729 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
 730         if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
 731                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
 732
 733                 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
 734                         /* Set the supported features */
 735                         if (ecmd.supported & SUPPORTED_10baseT_Half)
 736                                 desc->supported |= OFPPF_10MB_HD;
 737                         if (ecmd.supported & SUPPORTED_10baseT_Full)
 738                                 desc->supported |= OFPPF_10MB_FD;
 739                         if (ecmd.supported & SUPPORTED_100baseT_Half)
 740                                 desc->supported |= OFPPF_100MB_HD;
 741                         if (ecmd.supported & SUPPORTED_100baseT_Full)
 742                                 desc->supported |= OFPPF_100MB_FD;
 743                         if (ecmd.supported & SUPPORTED_1000baseT_Half)
 744                                 desc->supported |= OFPPF_1GB_HD;
 745                         if (ecmd.supported & SUPPORTED_1000baseT_Full)
 746                                 desc->supported |= OFPPF_1GB_FD;
 747                         if (ecmd.supported & SUPPORTED_10000baseT_Full)
 748                                 desc->supported |= OFPPF_10GB_FD;
 749                         if (ecmd.supported & SUPPORTED_TP)
 750                                 desc->supported |= OFPPF_COPPER;
 751                         if (ecmd.supported & SUPPORTED_FIBRE)
 752                                 desc->supported |= OFPPF_FIBER;
 753                         if (ecmd.supported & SUPPORTED_Autoneg)
 754                                 desc->supported |= OFPPF_AUTONEG;
 755 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 756                         if (ecmd.supported & SUPPORTED_Pause)
 757                                 desc->supported |= OFPPF_PAUSE;
 758                         if (ecmd.supported & SUPPORTED_Asym_Pause)
 759                                 desc->supported |= OFPPF_PAUSE_ASYM;
 760 #endif /* kernel >= 2.6.14 */
 761
 762                         /* Set the advertised features */
 763                         if (ecmd.advertising & ADVERTISED_10baseT_Half)
 764                                 desc->advertised |= OFPPF_10MB_HD;
 765                         if (ecmd.advertising & ADVERTISED_10baseT_Full)
 766                                 desc->advertised |= OFPPF_10MB_FD;
 767                         if (ecmd.advertising & ADVERTISED_100baseT_Half)
 768                                 desc->advertised |= OFPPF_100MB_HD;
 769                         if (ecmd.advertising & ADVERTISED_100baseT_Full)
 770                                 desc->advertised |= OFPPF_100MB_FD;
 771                         if (ecmd.advertising & ADVERTISED_1000baseT_Half)
 772                                 desc->advertised |= OFPPF_1GB_HD;
 773                         if (ecmd.advertising & ADVERTISED_1000baseT_Full)
 774                                 desc->advertised |= OFPPF_1GB_FD;
 775                         if (ecmd.advertising & ADVERTISED_10000baseT_Full)
 776                                 desc->advertised |= OFPPF_10GB_FD;
 777                         if (ecmd.advertising & ADVERTISED_TP)
 778                                 desc->advertised |= OFPPF_COPPER;
 779                         if (ecmd.advertising & ADVERTISED_FIBRE)
 780                                 desc->advertised |= OFPPF_FIBER;
 781                         if (ecmd.advertising & ADVERTISED_Autoneg)
 782                                 desc->advertised |= OFPPF_AUTONEG;
 783 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 784                         if (ecmd.advertising & ADVERTISED_Pause)
 785                                 desc->advertised |= OFPPF_PAUSE;
 786                         if (ecmd.advertising & ADVERTISED_Asym_Pause)
 787                                 desc->advertised |= OFPPF_PAUSE_ASYM;
 788 #endif /* kernel >= 2.6.14 */
 789
 790                         /* Set the current features */
 791                         if (ecmd.speed == SPEED_10)
 792                                 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
 793                         else if (ecmd.speed == SPEED_100)
 794                                 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
 795                         else if (ecmd.speed == SPEED_1000)
 796                                 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
 797                         else if (ecmd.speed == SPEED_10000)
 798                                 desc->curr = OFPPF_10GB_FD;
 799
 800                         if (ecmd.port == PORT_TP)
 801                                 desc->curr |= OFPPF_COPPER;
 802                         else if (ecmd.port == PORT_FIBRE)
 803                                 desc->curr |= OFPPF_FIBER;
 804
 805                         if (ecmd.autoneg)
 806                                 desc->curr |= OFPPF_AUTONEG;
 807                 }
 808         }
 809 #endif
 810         desc->curr = htonl(desc->curr);
 811         desc->supported = htonl(desc->supported);
 812         desc->advertised = htonl(desc->advertised);
 813         desc->peer = htonl(desc->peer);
 814 }
 815
 816 static int
 817 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
 818 {
 819         struct net_bridge_port *p;
 820         int port_count = 0;
 821
 822         ofr->datapath_id  = cpu_to_be64(dp->id);
 823
 824         ofr->n_buffers    = htonl(N_PKT_BUFFERS);
 825         ofr->n_tables     = dp->chain->n_tables;
 826         ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
 827         ofr->actions      = htonl(OFP_SUPPORTED_ACTIONS);
 828         memset(ofr->pad, 0, sizeof ofr->pad);
 829
 830         list_for_each_entry_rcu (p, &dp->port_list, node) {
 831                 fill_port_desc(p, &ofr->ports[port_count]);
 832                 port_count++;
 833         }
 834
 835         return port_count;
 836 }
 837
 838 int
 839 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
 840 {
 841         struct sk_buff *skb;
 842         struct ofp_switch_features *ofr;
 843         size_t ofr_len, port_max_len;
 844         int port_count;
 845
 846         /* Overallocate. */
 847         port_max_len = sizeof(struct ofp_phy_port) * DP_MAX_PORTS;
 848         ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
 849                                  OFPT_FEATURES_REPLY, sender, &skb);
 850         if (!ofr)
 851                 return -ENOMEM;
 852
 853         /* Fill. */
 854         port_count = fill_features_reply(dp, ofr);
 855
 856         /* Shrink to fit. */
 857         ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
 858         resize_openflow_skb(skb, &ofr->header, ofr_len);
 859         return send_openflow_skb(skb, sender);
 860 }
 861
 862 int
 863 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
 864 {
 865         struct sk_buff *skb;
 866         struct ofp_switch_config *osc;
 867
 868         osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
 869                                  &skb);
 870         if (!osc)
 871                 return -ENOMEM;
 872
 873         osc->flags = htons(dp->flags);
 874         osc->miss_send_len = htons(dp->miss_send_len);
 875
 876         return send_openflow_skb(skb, sender);
 877 }
 878
 879 int
 880 dp_send_hello(struct datapath *dp, const struct sender *sender,
 881               const struct ofp_header *request)
 882 {
 883         if (request->version < OFP_VERSION) {
 884                 char err[64];
 885                 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
 886                 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
 887                                   OFPHFC_INCOMPATIBLE, err, strlen(err));
 888                 return -EINVAL;
 889         } else {
 890                 struct sk_buff *skb;
 891                 struct ofp_header *reply;
 892
 893                 reply = alloc_openflow_skb(dp, sizeof *reply,
 894                                            OFPT_HELLO, sender, &skb);
 895                 if (!reply)
 896                         return -ENOMEM;
 897
 898                 return send_openflow_skb(skb, sender);
 899         }
 900 }
 901
 902 /* Callback function for a workqueue to disable an interface */
 903 static void
 904 down_port_cb(struct work_struct *work)
 905 {
 906         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 907                         port_task);
 908
 909         rtnl_lock();
 910         if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
 911                 if (net_ratelimit())
 912                         printk("problem bringing up port %s\n", p->dev->name);
 913         rtnl_unlock();
 914         p->config |= OFPPC_PORT_DOWN;
 915 }
 916
 917 /* Callback function for a workqueue to enable an interface */
 918 static void
 919 up_port_cb(struct work_struct *work)
 920 {
 921         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 922                         port_task);
 923
 924         rtnl_lock();
 925         if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
 926                 if (net_ratelimit())
 927                         printk("problem bringing down port %s\n", p->dev->name);
 928         rtnl_unlock();
 929         p->config &= ~OFPPC_PORT_DOWN;
 930 }
 931
 932 int
 933 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
 934 {
 935         unsigned long int flags;
 936         int port_no = ntohs(opm->port_no);
 937         struct net_bridge_port *p;
 938         p = (port_no < DP_MAX_PORTS ? dp->ports[port_no]
 939              : port_no == OFPP_LOCAL ? dp->local_port
 940              : NULL);
 941
 942         /* Make sure the port id hasn't changed since this was sent */
 943         if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
 944                 return -1;
 945
 946         spin_lock_irqsave(&p->lock, flags);
 947         if (opm->mask) {
 948                 uint32_t config_mask = ntohl(opm->mask);
 949                 p->config &= ~config_mask;
 950                 p->config |= ntohl(opm->config) & config_mask;
 951         }
 952
 953         /* Modifying the status of an interface requires taking a lock
 954          * that cannot be done from here.  For this reason, we use a shared
 955          * workqueue, which will cause it to be executed from a safer
 956          * context. */
 957         if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
 958                 if ((opm->config & htonl(OFPPC_PORT_DOWN))
 959                     && (p->config & OFPPC_PORT_DOWN) == 0) {
 960                         PREPARE_WORK(&p->port_task, down_port_cb);
 961                         schedule_work(&p->port_task);
 962                 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
 963                            && (p->config & OFPPC_PORT_DOWN)) {
 964                         PREPARE_WORK(&p->port_task, up_port_cb);
 965                         schedule_work(&p->port_task);
 966                 }
 967         }
 968         spin_unlock_irqrestore(&p->lock, flags);
 969
 970         return 0;
 971 }
 972
 973 /* Initialize the port status field of the bridge port. */
 974 static void
 975 init_port_status(struct net_bridge_port *p)
 976 {
 977         unsigned long int flags;
 978
 979         spin_lock_irqsave(&p->lock, flags);
 980
 981         if (p->dev->flags & IFF_UP)
 982                 p->config &= ~OFPPC_PORT_DOWN;
 983         else
 984                 p->config |= OFPPC_PORT_DOWN;
 985
 986         if (netif_carrier_ok(p->dev))
 987                 p->state &= ~OFPPS_LINK_DOWN;
 988         else
 989                 p->state |= OFPPS_LINK_DOWN;
 990
 991         spin_unlock_irqrestore(&p->lock, flags);
 992 }
 993
 994 int
 995 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
 996 {
 997         struct sk_buff *skb;
 998         struct ofp_port_status *ops;
 999
1000         ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
1001                                  &skb);
1002         if (!ops)
1003                 return -ENOMEM;
1004         ops->reason = status;
1005         memset(ops->pad, 0, sizeof ops->pad);
1006         fill_port_desc(p, &ops->desc);
1007
1008         return send_openflow_skb(skb, NULL);
1009 }
1010
1011 int
1012 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
1013                      enum ofp_flow_expired_reason reason)
1014 {
1015         struct sk_buff *skb;
1016         struct ofp_flow_expired *ofe;
1017
1018         if (!(dp->flags & OFPC_SEND_FLOW_EXP))
1019                 return 0;
1020
1021         ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
1022         if (!ofe)
1023                 return -ENOMEM;
1024
1025         flow_fill_match(&ofe->match, &flow->key);
1026
1027         ofe->priority = htons(flow->priority);
1028         ofe->reason = reason;
1029         memset(ofe->pad, 0, sizeof ofe->pad);
1030
1031         ofe->duration     = htonl((jiffies - flow->init_time) / HZ);
1032         memset(ofe->pad2, 0, sizeof ofe->pad2);
1033         ofe->packet_count = cpu_to_be64(flow->packet_count);
1034         ofe->byte_count   = cpu_to_be64(flow->byte_count);
1035
1036         return send_openflow_skb(skb, NULL);
1037 }
1038 EXPORT_SYMBOL(dp_send_flow_expired);
1039
1040 int
1041 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1042                 uint16_t type, uint16_t code, const void *data, size_t len)
1043 {
1044         struct sk_buff *skb;
1045         struct ofp_error_msg *oem;
1046
1047
1048         oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1049                         sender, &skb);
1050         if (!oem)
1051                 return -ENOMEM;
1052
1053         oem->type = htons(type);
1054         oem->code = htons(code);
1055         memcpy(oem->data, data, len);
1056
1057         return send_openflow_skb(skb, sender);
1058 }
1059
1060 int
1061 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1062                    const struct ofp_header *rq)
1063 {
1064         struct sk_buff *skb;
1065         struct ofp_header *reply;
1066
1067         reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1068                                    sender, &skb);
1069         if (!reply)
1070                 return -ENOMEM;
1071
1072         memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1073         return send_openflow_skb(skb, sender);
1074 }
1075
1076 /* Generic Netlink interface.
1077  *
1078  * See netlink(7) for an introduction to netlink.  See
1079  * http://linux-net.osdl.org/index.php/Netlink for more information and
1080  * pointers on how to work with netlink and Generic Netlink in the kernel and
1081  * in userspace. */
1082
1083 static struct genl_family dp_genl_family = {
1084         .id = GENL_ID_GENERATE,
1085         .hdrsize = 0,
1086         .name = DP_GENL_FAMILY_NAME,
1087         .version = 1,
1088         .maxattr = DP_GENL_A_MAX,
1089 };
1090
1091 /* Attribute policy: what each attribute may contain.  */
1092 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1093         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1094         [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1095         [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1096 };
1097
1098 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1099 {
1100         if (!info->attrs[DP_GENL_A_DP_IDX])
1101                 return -EINVAL;
1102
1103         return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1104 }
1105
1106 static struct genl_ops dp_genl_ops_add_dp = {
1107         .cmd = DP_GENL_C_ADD_DP,
1108         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1109         .policy = dp_genl_policy,
1110         .doit = dp_genl_add,
1111         .dumpit = NULL,
1112 };
1113
1114 struct datapath *dp_get(int dp_idx)
1115 {
1116         if (dp_idx < 0 || dp_idx > DP_MAX)
1117                 return NULL;
1118         return rcu_dereference(dps[dp_idx]);
1119 }
1120
1121 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1122 {
1123         struct datapath *dp;
1124         int err;
1125
1126         if (!info->attrs[DP_GENL_A_DP_IDX])
1127                 return -EINVAL;
1128
1129         dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
1130         if (!dp)
1131                 err = -ENOENT;
1132         else {
1133                 del_dp(dp);
1134                 err = 0;
1135         }
1136         return err;
1137 }
1138
1139 static struct genl_ops dp_genl_ops_del_dp = {
1140         .cmd = DP_GENL_C_DEL_DP,
1141         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1142         .policy = dp_genl_policy,
1143         .doit = dp_genl_del,
1144         .dumpit = NULL,
1145 };
1146
1147 /* Queries a datapath for related information.  Currently the only relevant
1148  * information is the datapath's multicast group ID.  Really we want one
1149  * multicast group per datapath, but because of locking issues[*] we can't
1150  * easily get one.  Thus, every datapath will currently return the same
1151  * global multicast group ID, but in the future it would be nice to fix that.
1152  *
1153  * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1154  *       mutex, and genl_register_mc_group, called to acquire a new multicast
1155  *       group ID, also acquires genl_lock, thus deadlock.
1156  */
1157 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1158 {
1159         struct datapath *dp;
1160         struct sk_buff *ans_skb = NULL;
1161         int dp_idx;
1162         int err = -ENOMEM;
1163
1164         if (!info->attrs[DP_GENL_A_DP_IDX])
1165                 return -EINVAL;
1166
1167         rcu_read_lock();
1168         dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1169         dp = dp_get(dp_idx);
1170         if (!dp)
1171                 err = -ENOENT;
1172         else {
1173                 void *data;
1174                 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1175                 if (!ans_skb) {
1176                         err = -ENOMEM;
1177                         goto err;
1178                 }
1179                 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1180                                          0, DP_GENL_C_QUERY_DP);
1181                 if (data == NULL) {
1182                         err = -ENOMEM;
1183                         goto err;
1184                 }
1185                 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1186                 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1187
1188                 genlmsg_end(ans_skb, data);
1189                 err = genlmsg_reply(ans_skb, info);
1190                 if (!err)
1191                         ans_skb = NULL;
1192         }
1193 err:
1194 nla_put_failure:
1195         if (ans_skb)
1196                 kfree_skb(ans_skb);
1197         rcu_read_unlock();
1198         return err;
1199 }
1200
1201 static struct genl_ops dp_genl_ops_query_dp = {
1202         .cmd = DP_GENL_C_QUERY_DP,
1203         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1204         .policy = dp_genl_policy,
1205         .doit = dp_genl_query,
1206         .dumpit = NULL,
1207 };
1208
1209 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1210 {
1211         struct datapath *dp;
1212         struct net_device *port;
1213         int err;
1214
1215         if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1216                 return -EINVAL;
1217
1218         /* Get datapath. */
1219         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1220         if (!dp) {
1221                 err = -ENOENT;
1222                 goto out;
1223         }
1224
1225         /* Get interface to add/remove. */
1226         port = dev_get_by_name(&init_net,
1227                         nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1228         if (!port) {
1229                 err = -ENOENT;
1230                 goto out;
1231         }
1232
1233         /* Execute operation. */
1234         if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1235                 err = add_switch_port(dp, port);
1236         else {
1237                 if (port->br_port == NULL || port->br_port->dp != dp) {
1238                         err = -ENOENT;
1239                         goto out_put;
1240                 }
1241                 err = dp_del_switch_port(port->br_port);
1242         }
1243
1244 out_put:
1245         dev_put(port);
1246 out:
1247         return err;
1248 }
1249
1250 static struct genl_ops dp_genl_ops_add_port = {
1251         .cmd = DP_GENL_C_ADD_PORT,
1252         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1253         .policy = dp_genl_policy,
1254         .doit = dp_genl_add_del_port,
1255         .dumpit = NULL,
1256 };
1257
1258 static struct genl_ops dp_genl_ops_del_port = {
1259         .cmd = DP_GENL_C_DEL_PORT,
1260         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1261         .policy = dp_genl_policy,
1262         .doit = dp_genl_add_del_port,
1263         .dumpit = NULL,
1264 };
1265
1266 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1267 {
1268         struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1269         struct datapath *dp;
1270         struct ofp_header *oh;
1271         struct sender sender;
1272         int err;
1273
1274         if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1275                 return -EINVAL;
1276
1277         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1278         if (!dp)
1279                 return -ENOENT;
1280
1281         if (nla_len(va) < sizeof(struct ofp_header))
1282                 return -EINVAL;
1283         oh = nla_data(va);
1284
1285         sender.xid = oh->xid;
1286         sender.pid = info->snd_pid;
1287         sender.seq = info->snd_seq;
1288
1289         mutex_lock(&dp_mutex);
1290         err = fwd_control_input(dp->chain, &sender,
1291                                 nla_data(va), nla_len(va));
1292         mutex_unlock(&dp_mutex);
1293         return err;
1294 }
1295
1296 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1297         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1298 };
1299
1300 static int desc_stats_dump(struct datapath *dp, void *state,
1301                             void *body, int *body_len)
1302 {
1303         struct ofp_desc_stats *ods = body;
1304         int n_bytes = sizeof *ods;
1305
1306         if (n_bytes > *body_len) {
1307                 return -ENOBUFS;
1308         }
1309         *body_len = n_bytes;
1310
1311         strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1312         strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1313         strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1314         strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1315
1316         return 0;
1317 }
1318
1319 struct flow_stats_state {
1320         int table_idx;
1321         struct sw_table_position position;
1322         const struct ofp_flow_stats_request *rq;
1323
1324         void *body;
1325         int bytes_used, bytes_allocated;
1326 };
1327
1328 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1329                            void **state)
1330 {
1331         const struct ofp_flow_stats_request *fsr = body;
1332         struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1333         if (!s)
1334                 return -ENOMEM;
1335         s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1336         memset(&s->position, 0, sizeof s->position);
1337         s->rq = fsr;
1338         *state = s;
1339         return 0;
1340 }
1341
1342 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1343 {
1344         struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1345         struct flow_stats_state *s = private;
1346         struct ofp_flow_stats *ofs;
1347         int length;
1348
1349         length = sizeof *ofs + sf_acts->actions_len;
1350         if (length + s->bytes_used > s->bytes_allocated)
1351                 return 1;
1352
1353         ofs = s->body + s->bytes_used;
1354         ofs->length          = htons(length);
1355         ofs->table_id        = s->table_idx;
1356         ofs->pad             = 0;
1357         ofs->match.wildcards = htonl(flow->key.wildcards);
1358         ofs->match.in_port   = flow->key.in_port;
1359         memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1360         memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1361         ofs->match.dl_vlan   = flow->key.dl_vlan;
1362         ofs->match.dl_type   = flow->key.dl_type;
1363         ofs->match.nw_src    = flow->key.nw_src;
1364         ofs->match.nw_dst    = flow->key.nw_dst;
1365         ofs->match.nw_proto  = flow->key.nw_proto;
1366         ofs->match.pad       = 0;
1367         ofs->match.tp_src    = flow->key.tp_src;
1368         ofs->match.tp_dst    = flow->key.tp_dst;
1369         ofs->duration        = htonl((jiffies - flow->init_time) / HZ);
1370         ofs->priority        = htons(flow->priority);
1371         ofs->idle_timeout    = htons(flow->idle_timeout);
1372         ofs->hard_timeout    = htons(flow->hard_timeout);
1373         memset(ofs->pad2, 0, sizeof ofs->pad2);
1374         ofs->packet_count    = cpu_to_be64(flow->packet_count);
1375         ofs->byte_count      = cpu_to_be64(flow->byte_count);
1376         memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1377
1378         s->bytes_used += length;
1379         return 0;
1380 }
1381
1382 static int flow_stats_dump(struct datapath *dp, void *state,
1383                            void *body, int *body_len)
1384 {
1385         struct flow_stats_state *s = state;
1386         struct sw_flow_key match_key;
1387         int error = 0;
1388
1389         s->bytes_used = 0;
1390         s->bytes_allocated = *body_len;
1391         s->body = body;
1392
1393         flow_extract_match(&match_key, &s->rq->match);
1394         while (s->table_idx < dp->chain->n_tables
1395                && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1396         {
1397                 struct sw_table *table = dp->chain->tables[s->table_idx];
1398
1399                 error = table->iterate(table, &match_key, &s->position,
1400                                        flow_stats_dump_callback, s);
1401                 if (error)
1402                         break;
1403
1404                 s->table_idx++;
1405                 memset(&s->position, 0, sizeof s->position);
1406         }
1407         *body_len = s->bytes_used;
1408
1409         /* If error is 0, we're done.
1410          * Otherwise, if some bytes were used, there are more flows to come.
1411          * Otherwise, we were not able to fit even a single flow in the body,
1412          * which indicates that we have a single flow with too many actions to
1413          * fit.  We won't ever make any progress at that rate, so give up. */
1414         return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1415 }
1416
1417 static void flow_stats_done(void *state)
1418 {
1419         kfree(state);
1420 }
1421
1422 static int aggregate_stats_init(struct datapath *dp,
1423                                 const void *body, int body_len,
1424                                 void **state)
1425 {
1426         *state = (void *)body;
1427         return 0;
1428 }
1429
1430 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1431 {
1432         struct ofp_aggregate_stats_reply *rpy = private;
1433         rpy->packet_count += flow->packet_count;
1434         rpy->byte_count += flow->byte_count;
1435         rpy->flow_count++;
1436         return 0;
1437 }
1438
1439 static int aggregate_stats_dump(struct datapath *dp, void *state,
1440                                 void *body, int *body_len)
1441 {
1442         struct ofp_aggregate_stats_request *rq = state;
1443         struct ofp_aggregate_stats_reply *rpy;
1444         struct sw_table_position position;
1445         struct sw_flow_key match_key;
1446         int table_idx;
1447
1448         if (*body_len < sizeof *rpy)
1449                 return -ENOBUFS;
1450         rpy = body;
1451         *body_len = sizeof *rpy;
1452
1453         memset(rpy, 0, sizeof *rpy);
1454
1455         flow_extract_match(&match_key, &rq->match);
1456         table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1457         memset(&position, 0, sizeof position);
1458         while (table_idx < dp->chain->n_tables
1459                && (rq->table_id == 0xff || rq->table_id == table_idx))
1460         {
1461                 struct sw_table *table = dp->chain->tables[table_idx];
1462                 int error;
1463
1464                 error = table->iterate(table, &match_key, &position,
1465                                        aggregate_stats_dump_callback, rpy);
1466                 if (error)
1467                         return error;
1468
1469                 table_idx++;
1470                 memset(&position, 0, sizeof position);
1471         }
1472
1473         rpy->packet_count = cpu_to_be64(rpy->packet_count);
1474         rpy->byte_count = cpu_to_be64(rpy->byte_count);
1475         rpy->flow_count = htonl(rpy->flow_count);
1476         return 0;
1477 }
1478
1479 static int table_stats_dump(struct datapath *dp, void *state,
1480                             void *body, int *body_len)
1481 {
1482         struct ofp_table_stats *ots;
1483         int n_bytes = dp->chain->n_tables * sizeof *ots;
1484         int i;
1485         if (n_bytes > *body_len)
1486                 return -ENOBUFS;
1487         *body_len = n_bytes;
1488         for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1489                 struct sw_table_stats stats;
1490                 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1491                 strncpy(ots->name, stats.name, sizeof ots->name);
1492                 ots->table_id = i;
1493                 ots->wildcards = htonl(stats.wildcards);
1494                 memset(ots->pad, 0, sizeof ots->pad);
1495                 ots->max_entries = htonl(stats.max_flows);
1496                 ots->active_count = htonl(stats.n_flows);
1497                 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1498                 ots->matched_count = cpu_to_be64(stats.n_matched);
1499         }
1500         return 0;
1501 }
1502
1503 struct port_stats_state {
1504         int port;
1505 };
1506
1507 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1508                            void **state)
1509 {
1510         struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1511         if (!s)
1512                 return -ENOMEM;
1513         s->port = 0;
1514         *state = s;
1515         return 0;
1516 }
1517
1518 static int port_stats_dump(struct datapath *dp, void *state,
1519                            void *body, int *body_len)
1520 {
1521         struct port_stats_state *s = state;
1522         struct ofp_port_stats *ops;
1523         int n_ports, max_ports;
1524         int i;
1525
1526         max_ports = *body_len / sizeof *ops;
1527         if (!max_ports)
1528                 return -ENOMEM;
1529         ops = body;
1530
1531         n_ports = 0;
1532         for (i = s->port; i < DP_MAX_PORTS && n_ports < max_ports; i++) {
1533                 struct net_bridge_port *p = dp->ports[i];
1534                 struct net_device_stats *stats;
1535                 if (!p)
1536                         continue;
1537                 stats = p->dev->get_stats(p->dev);
1538                 ops->port_no = htons(p->port_no);
1539                 memset(ops->pad, 0, sizeof ops->pad);
1540                 ops->rx_packets   = cpu_to_be64(stats->rx_packets);
1541                 ops->tx_packets   = cpu_to_be64(stats->tx_packets);
1542                 ops->rx_bytes     = cpu_to_be64(stats->rx_bytes);
1543                 ops->tx_bytes     = cpu_to_be64(stats->tx_bytes);
1544                 ops->rx_dropped   = cpu_to_be64(stats->rx_dropped);
1545                 ops->tx_dropped   = cpu_to_be64(stats->tx_dropped);
1546                 ops->rx_errors    = cpu_to_be64(stats->rx_errors);
1547                 ops->tx_errors    = cpu_to_be64(stats->tx_errors);
1548                 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1549                 ops->rx_over_err  = cpu_to_be64(stats->rx_over_errors);
1550                 ops->rx_crc_err   = cpu_to_be64(stats->rx_crc_errors);
1551                 ops->collisions   = cpu_to_be64(stats->collisions);
1552                 n_ports++;
1553                 ops++;
1554         }
1555         s->port = i;
1556         *body_len = n_ports * sizeof *ops;
1557         return n_ports >= max_ports;
1558 }
1559
1560 static void port_stats_done(void *state)
1561 {
1562         kfree(state);
1563 }
1564
1565 struct stats_type {
1566         /* Minimum and maximum acceptable number of bytes in body member of
1567          * struct ofp_stats_request. */
1568         size_t min_body, max_body;
1569
1570         /* Prepares to dump some kind of statistics on 'dp'.  'body' and
1571          * 'body_len' are the 'body' member of the struct ofp_stats_request.
1572          * Returns zero if successful, otherwise a negative error code.
1573          * May initialize '*state' to state information.  May be null if no
1574          * initialization is required.*/
1575         int (*init)(struct datapath *dp, const void *body, int body_len,
1576                     void **state);
1577
1578         /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1579          * modifies '*body_len' to reflect the number of bytes actually used.
1580          * ('body' will be transmitted as the 'body' member of struct
1581          * ofp_stats_reply.) */
1582         int (*dump)(struct datapath *dp, void *state,
1583                     void *body, int *body_len);
1584
1585         /* Cleans any state created by the init or dump functions.  May be null
1586          * if no cleanup is required. */
1587         void (*done)(void *state);
1588 };
1589
1590 static const struct stats_type stats[] = {
1591         [OFPST_DESC] = {
1592                 0,
1593                 0,
1594                 NULL,
1595                 desc_stats_dump,
1596                 NULL
1597         },
1598         [OFPST_FLOW] = {
1599                 sizeof(struct ofp_flow_stats_request),
1600                 sizeof(struct ofp_flow_stats_request),
1601                 flow_stats_init,
1602                 flow_stats_dump,
1603                 flow_stats_done
1604         },
1605         [OFPST_AGGREGATE] = {
1606                 sizeof(struct ofp_aggregate_stats_request),
1607                 sizeof(struct ofp_aggregate_stats_request),
1608                 aggregate_stats_init,
1609                 aggregate_stats_dump,
1610                 NULL
1611         },
1612         [OFPST_TABLE] = {
1613                 0,
1614                 0,
1615                 NULL,
1616                 table_stats_dump,
1617                 NULL
1618         },
1619         [OFPST_PORT] = {
1620                 0,
1621                 0,
1622                 port_stats_init,
1623                 port_stats_dump,
1624                 port_stats_done
1625         },
1626 };
1627
1628 static int
1629 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1630 {
1631         struct datapath *dp;
1632         struct sender sender;
1633         const struct stats_type *s;
1634         struct ofp_stats_reply *osr;
1635         int dp_idx;
1636         int max_openflow_len, body_len;
1637         void *body;
1638         int err;
1639
1640         /* Set up the cleanup function for this dump.  Linux 2.6.20 and later
1641          * support setting up cleanup functions via the .doneit member of
1642          * struct genl_ops.  This kluge supports earlier versions also. */
1643         cb->done = dp_genl_openflow_done;
1644
1645         sender.pid = NETLINK_CB(cb->skb).pid;
1646         sender.seq = cb->nlh->nlmsg_seq;
1647         if (!cb->args[0]) {
1648                 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1649                 struct ofp_stats_request *rq;
1650                 struct nlattr *va;
1651                 size_t len, body_len;
1652                 int type;
1653
1654                 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1655                                   dp_genl_openflow_policy);
1656                 if (err < 0)
1657                         return err;
1658
1659                 if (!attrs[DP_GENL_A_DP_IDX])
1660                         return -EINVAL;
1661                 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1662                 dp = dp_get(dp_idx);
1663                 if (!dp)
1664                         return -ENOENT;
1665
1666                 va = attrs[DP_GENL_A_OPENFLOW];
1667                 len = nla_len(va);
1668                 if (!va || len < sizeof *rq)
1669                         return -EINVAL;
1670
1671                 rq = nla_data(va);
1672                 sender.xid = rq->header.xid;
1673                 type = ntohs(rq->type);
1674                 if (rq->header.version != OFP_VERSION) {
1675                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1676                                           OFPBRC_BAD_VERSION, rq, len);
1677                         return -EINVAL;
1678                 }
1679                 if (rq->header.type != OFPT_STATS_REQUEST
1680                     || ntohs(rq->header.length) != len)
1681                         return -EINVAL;
1682
1683                 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1684                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1685                                           OFPBRC_BAD_STAT, rq, len);
1686                         return -EINVAL;
1687                 }
1688
1689                 s = &stats[type];
1690                 body_len = len - offsetof(struct ofp_stats_request, body);
1691                 if (body_len < s->min_body || body_len > s->max_body)
1692                         return -EINVAL;
1693
1694                 cb->args[0] = 1;
1695                 cb->args[1] = dp_idx;
1696                 cb->args[2] = type;
1697                 cb->args[3] = rq->header.xid;
1698                 if (s->init) {
1699                         void *state;
1700                         err = s->init(dp, rq->body, body_len, &state);
1701                         if (err)
1702                                 return err;
1703                         cb->args[4] = (long) state;
1704                 }
1705         } else if (cb->args[0] == 1) {
1706                 sender.xid = cb->args[3];
1707                 dp_idx = cb->args[1];
1708                 s = &stats[cb->args[2]];
1709
1710                 dp = dp_get(dp_idx);
1711                 if (!dp)
1712                         return -ENOENT;
1713         } else {
1714                 return 0;
1715         }
1716
1717         osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1718                                    &max_openflow_len);
1719         if (IS_ERR(osr))
1720                 return PTR_ERR(osr);
1721         osr->type = htons(s - stats);
1722         osr->flags = 0;
1723         resize_openflow_skb(skb, &osr->header, max_openflow_len);
1724         body = osr->body;
1725         body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1726
1727         err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1728         if (err >= 0) {
1729                 if (!err)
1730                         cb->args[0] = 2;
1731                 else
1732                         osr->flags = ntohs(OFPSF_REPLY_MORE);
1733                 resize_openflow_skb(skb, &osr->header,
1734                                     (offsetof(struct ofp_stats_reply, body)
1735                                      + body_len));
1736                 err = skb->len;
1737         }
1738
1739         return err;
1740 }
1741
1742 static int
1743 dp_genl_openflow_done(struct netlink_callback *cb)
1744 {
1745         if (cb->args[0]) {
1746                 const struct stats_type *s = &stats[cb->args[2]];
1747                 if (s->done)
1748                         s->done((void *) cb->args[4]);
1749         }
1750         return 0;
1751 }
1752
1753 static struct genl_ops dp_genl_ops_openflow = {
1754         .cmd = DP_GENL_C_OPENFLOW,
1755         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1756         .policy = dp_genl_openflow_policy,
1757         .doit = dp_genl_openflow,
1758         .dumpit = dp_genl_openflow_dumpit,
1759 };
1760
1761 static struct genl_ops *dp_genl_all_ops[] = {
1762         /* Keep this operation first.  Generic Netlink dispatching
1763          * looks up operations with linear search, so we want it at the
1764          * front. */
1765         &dp_genl_ops_openflow,
1766
1767         &dp_genl_ops_add_dp,
1768         &dp_genl_ops_del_dp,
1769         &dp_genl_ops_query_dp,
1770         &dp_genl_ops_add_port,
1771         &dp_genl_ops_del_port,
1772 };
1773
1774 static int dp_init_netlink(void)
1775 {
1776         int err;
1777         int i;
1778
1779         err = genl_register_family(&dp_genl_family);
1780         if (err)
1781                 return err;
1782
1783         for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1784                 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1785                 if (err)
1786                         goto err_unregister;
1787         }
1788
1789         strcpy(mc_group.name, "openflow");
1790         err = genl_register_mc_group(&dp_genl_family, &mc_group);
1791         if (err < 0)
1792                 goto err_unregister;
1793
1794         return 0;
1795
1796 err_unregister:
1797         genl_unregister_family(&dp_genl_family);
1798                 return err;
1799 }
1800
1801 static void dp_uninit_netlink(void)
1802 {
1803         genl_unregister_family(&dp_genl_family);
1804 }
1805
1806 static int __init dp_init(void)
1807 {
1808         int err;
1809
1810         printk("OpenFlow "VERSION", built "__DATE__" "__TIME__", "
1811                "protocol 0x%02x\n", OFP_VERSION);
1812
1813         err = flow_init();
1814         if (err)
1815                 goto error;
1816
1817         err = register_netdevice_notifier(&dp_device_notifier);
1818         if (err)
1819                 goto error_flow_exit;
1820
1821         err = dp_init_netlink();
1822         if (err)
1823                 goto error_unreg_notifier;
1824
1825         /* Hook into callback used by the bridge to intercept packets.
1826          * Parasites we are. */
1827         if (br_handle_frame_hook)
1828                 printk("openflow: hijacking bridge hook\n");
1829         br_handle_frame_hook = dp_frame_hook;
1830
1831         return 0;
1832
1833 error_unreg_notifier:
1834         unregister_netdevice_notifier(&dp_device_notifier);
1835 error_flow_exit:
1836         flow_exit();
1837 error:
1838         printk(KERN_EMERG "openflow: failed to install!");
1839         return err;
1840 }
1841
1842 static void dp_cleanup(void)
1843 {
1844         fwd_exit();
1845         dp_uninit_netlink();
1846         unregister_netdevice_notifier(&dp_device_notifier);
1847         flow_exit();
1848         br_handle_frame_hook = NULL;
1849 }
1850
1851 module_init(dp_init);
1852 module_exit(dp_cleanup);
1853
1854 MODULE_DESCRIPTION("OpenFlow switching datapath");
1855 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1856 MODULE_LICENSE("GPL");